Since pertinent information was withheld (that it didn't know), then by your own post you acknowledge it was a lie of omission.

The stupidity of people these days is truly beyond belief. And, yes, get the f off my lawn.

We learned back in the 80s that trying to get a neural net to emphasise what you want is actually very difficult. What it will tend to emphasise are the assumptions that underly the test data, and that's usually a completely different sort of fiction.

But was that figure provided by AI?

Even if not, we all know that 793% of all statistics are invented.

If something is inaccurately presented as being the truth, then it is a lie of omission because it is dishonest about the fact that the information isn't actually known.

Gemini is exceptionally bad, as LLMs go. I really have no idea why it is so dreadful, even compared to other LLMs. It isn't context window. and it doesn't seem to be training material either.

Cyber Implications have been noted. Mondas security is to be Cyber Vibed until we have Cyber Security capable of defeating The Doctor.

When I test the different AI systems, Google's AI system loses track of complex problems incredibly quickly. It's great on simple stuff, but for complex stuff, it's useless.

Unfortunately.... advice, overviews, etc, are very very complex problems indeed, which means that you're hitting the weakspot of their system.

I've designed a few machines - some rather more insane than others - in meticulous detail using AI. What I have not done, so far, is get an engineer to review the designs to see if any of them can be turned into something that would be usable. My suspicion is that a few might be made workable, but that has to be verified.

Having said that, producing the design probably took a significant amount of compute power and a significant amount of water. If I'd fermented that same quantity of water and provided wine to an engineering team that cost the same as the computing resources consumed, I'd probably have better designs.But, that too, is unverified. As before, it's perfectly verifiable, it just hasn't been so far.

If an engineer looks at the design and dies laughing, then I'm probably liable for funeral costs but at least there would be absolutely no question as to how good AI is at challenging engineering concepts. On the other hand, if they pause and say that there's actually a neat idea in a few of the concepts, then it becomes a question of how much of that was ideas I put in and how much is stuff the AI actually put together. Again, though, we'd have a metric.

That, to me, is the crux. It's all fine and well arguing over whether AI is any good or not (and, tbh, I would say that my feeling is that you're absolutely right), but this should be definitively measured and quantified, not assumed. There may be far better benchmarks than the designs I have - I'm good but I'm not one of the greats, so the odds of someone coming up with better measures seems high. But we're not seeing those, we're just seeing toy tests by journalists and that's not a good measure of real-world usability.

If no such benchmark values actually appear, then I think it's fair to argue that it's because nobody believes any AI out there is going to do well at them.

(I can tell you now, Gemini won't. Gemini is next to useless -- but on the Other Side.)

This means you shoud NOT, under any circumstance, run Claude at 88mph. Unless you really want to.

According to the writeup; there are two methods: it is possible for an extension to mark some parts of itself as 'web accessible'; and linkedin has assembled at least one characteristic file for 6,1000-odd extension IDs and attempts to fetch it to confirm/deny the extension's presence.

The other is based on the fact that the whole point of many extensions is to modify the site in some way; but the site normally has largely unfettered access to inspect itself, so they have theirs set up to walk the entire DOM looking for any references to "chrome-extension://" and snagging the IDs if found.

Not exactly a 'declare installed extensions'; but it looks like, out of some combination of supporting the use cases where an extension and page actively interact by design and either not wanting the possibility or not wanting the complexity of trying to enable 'invisible' edits(presumably some sort of 'shadow' DOM mechanism where as far as the site and everything delivered with it knows only its unedited DOM and resources exist; but the one the user sees is an extension-modified copy of that one, which sounds like it could get messy), inferential attacks are fairly easy and powerful.

Where exactly does supporting 3 people on $133k/year count as 'upper middle class'? You could be doing a lot worse, and many are; but that's not just tons of money in a HCOL area; and that's also lower than twice the median salary for full time employees with bachelor's degrees; so you are calling either a single income household doing a bit better than median or a dual income one doing worse 'upper middle class'; which seems pretty ambitious.

It's hard to train a cow to keep its cellphone charged; so you can see why you need something to keep their tracking collars topped up.

Realistically, the status quo has arguably outrun the dystopia there. Your phone already does far more than anything you could get into the power envelope of a bracelet or embedded chip implant, and if for some reason you've raised enough eyebrows that you'd be hauled in for an RFID read DNA is a pretty indelible identifier.

It's not 100% ironclad; but penetration is broad enough that you've basically got the majority carrying highly fingerprintable RF beacons and the minority standing out for their relative radio silence and attempts to deal in cash. Expensive and uncomfortable ankle trackers are good business and feel nice and punitive, just to remind the wrong sort of people we aren't happy with them; but you don't really need to impose a surveillance society when it will build itself for you.

My (admittedly anecdotal from the totally unscientific sample of random stuff I've had reason to work on) impression is that some 'shared' BMC ports had oddities related to network controller sideband interface speeds, since NC-SI is what the BMC is depending on if the NIC is on someone else's PCIe root. It's not like the BMC actually needs a faster link for much(normal management traffic probably doesn't fill 10mb and mounting virtual media may be literally once-in-a-lifetime) so the actual speed of the NC-SI interface was not a burning priority; but it left things up to the NIC whether it would support remaining at gigabit speeds and just quietly slipping the trickle of shared traffic in(presumably slightly more complex; but seems to be what the newer ones do) or if it would knock the link rate down visibly to simplify the case.

You see little echoes of similar behavior elsewhere. The intel desktop and laptop NICs that support 'vPRO' will be GB or 2.5GB when the system is on; but quietly drop back to 10 or 10/100 when it is off and it's just the management engine listening. Some enterprise vendor USB docks do similar things; looks like a normal USB NIC when the OS is up; but drops to a lower speed and operates quietly over, I think, some sort of oddball vendor-defined messages if one of their systems is plugged in but off.

I was hoping at the bottom of the article it would say that Professor Utonium accidentally added Chemical X.